PAPERS OF THE 131ST ASA ANNUAL MEETING

Predicting Risk for Serious Complications With Bariatric Surgery

Results from the Michigan Bariatric Surgery Collaborative
Jonathan F. Finks, MD, Kerry L. Kole, DO, Panduranga R. Yenumula, MD, Wayne J. English, MD,
Kevin R. Krause, MD, Arthur M. Carlin, MD, Jeffrey A. Genaw, MD, Mousumi Banerjee, PhD,
John D. Birkmeyer, MD, and Nancy J. Birkmeyer, PhD*; for the Michigan Bariatric Surgery Collaborative,
from the Center for Healthcare Outcomes and Policy

betes, cardiovascular disease and other health problems after obesity

Objectives: To develop a risk prediction model for serious complications after
surgery.13 Several studies have also documented a survival benefit
bariatric surgery.
for obese patients undergoing bariatric surgery.46 Given these re-
Background: Despite evidence for improved safety with bariatric surgery,
sults, rates of bariatric surgery have increased dramatically over the
serious complications remain a concern for patients, providers and payers.
last decade, with over 220,000 cases reported in the United States and
There is little population-level data on which risk factors can be used to
Canada in 2008.7,8 Although recent evidence suggests a trend toward
identify patients at high risk for major morbidity.
improved safety with bariatric surgery,9 rates of serious complica-
Methods: The Michigan Bariatric Surgery Collaborative is a statewide con-
tions, ranging between 2.5% and 4%,1012 remain a significant concern
sortium of hospitals and surgeons, which maintains an externally-audited
to patients, providers and payers. These events can lead to prolonged
prospective clinical registry. We analyzed data from 25,469 patients under-
hospitalization, lasting disability and significantly increased costs.
going bariatric surgery between June 2006 and December 2010. Significant
There would be substantial benefit in identifying factors which
risk factors on univariable analysis were entered into a multivariable logistic
predicted an increased risk for serious complications among patients
regression model to identify factors associated with serious complications
undergoing bariatric surgery. The ability to predict surgical risk would
(life threatening and/or associated with lasting disability) within 30 days of
help with patient selection and targeting of risk factors which could
surgery. Bootstrap resampling was performed to obtain bias-corrected confi-
be improved upon preoperatively. Furthermore, better tools for risk
dence intervals and c-statistic.
prediction would improve the process of informed consent before
Results: Overall, 644 patients (2.5%) experienced a serious complication.
surgery and may help with selecting the appropriate procedure for
Significant risk factors (P < 0.05) included: prior VTE (odds ratio [OR]
an individual patient. An accurate risk prediction model would help
1.90, confidence interval [CI] 1.412.54); mobility limitations (OR 1.61, CI
with risk stratification across studies of bariatric surgery and would
1.232.13); coronary artery disease (OR 1.53, CI 1.172.02); age over 50
provide a foundation for evidence-based insurance approvals for these
(OR 1.38, CI 1.181.61); pulmonary disease (OR 1.37, CI 1.151.64); male
procedures.
gender (OR 1.26, CI 1.061.50); smoking history (OR 1.20, CI 1.021.40);
In this context, we performed a risk-adjusted analysis from
and procedure type (reference lap band): duodenal switch (OR 9.68, CI 6.05
a prospective, statewide clinical registry to identify significant risk
15.49); laparoscopic gastric bypass (OR 3.58, CI 2.794.64); open gastric
factors for serious complications after bariatric surgery and developed
bypass (OR 3.51, CI 2.385.22); sleeve gastrectomy (OR 2.46, CI 1.733.50).
a preoperative risk calculator based on those factors.
The c-statistic was 0.68 (bias-corrected to 0.66) and the model was well-
calibrated across deciles of predicted risk.
Conclusions: We have developed and validated a population-based risk scor- METHODS
ing system for serious complications after bariatric surgery. We expect that Study Sample
this scoring system will improve the process of informed consent, facilitate Our study is based on analysis of data from the Michigan
the selection of procedures for high-risk patients, and allow for better risk Bariatric Surgery Collaborative (MBSC), a payer-funded quality im-
stratification across studies of bariatric surgery. provement program that administers a prospective, externally audited
(Ann Surg 2011;254:633640) clinical outcomes registry of patients undergoing bariatric surgery in
Michigan. The MBSC, a consortium of 29 Michigan hospitals and 75

B ariatric surgery has emerged as a highly effective treatment for se-

vere obesity and weight-related comorbidities. Numerous studies
have demonstrated sustained weight loss and improvements in dia-
From the *Department of Surgery, University of Michigan, Ann Arbor, MI;
Department of Surgery, St. John Macomb-Oakland Hospital, Madison
Heights, MI; Department of Surgery, Michigan State University, Lansing,
MI; Department of Surgery, Marquette General Hospital, Marquette, MI;
Department of Surgery, William Beaumont Hospital, Royal Oak, MI; **De-
partment of Surgery, Henry Ford Hospital, Detroit, MI; and Department of
Biostatistics, University of Michigan, Ann Arbor, MI.
Disclosure: Supported by a longitudinal quality improvement contract from Blue
Cross and Blue Shield of Michigan and Blue Care Network, and funding from
the Agency for Healthcare Research and Quality (1 R01 HS018050-01A1) [Dr.
N. Birkmeyer].
Reprints: Jonathan F. Finks, MD, 2210F Taubman Center, 1500 E. Medical Center
Drive, Ann Arbor, MI 48109, E-mail: jfinks@med.umich.edu.
Copyright C 2011 by Lippincott Williams & Wilkins
ISSN: 0003-4932/11/25404-0633
DOI: 10.1097/SLA.0b013e318230058c
surgeons performing bariatric surgery, has been described in detail
elsewhere.13 Participation in the MBSC is voluntary and any hospital
that performs at least 25 bariatric procedures per year is eligible to
participate. The MBSC currently enrolls about 6000 patients annu-
ally into its clinical registry. Participating hospitals submit data on all
patients undergoing primary and revisional bariatric surgery. Proce-
dures meeting this definition include open and laparoscopic gastric
bypass, adjustable gastric banding, sleeve gastrectomy, and biliopan-
creatic diversion with duodenal switch. The registry also includes
bariatric operations performed on an outpatient basis.
For the clinical registry, patient data are obtained from chart
abstraction at the end of the perioperative period (in-hospital and up to
30 days after surgery), and include information on preoperative clin-
ical characteristics and comorbid conditions, as well as perioperative
processes of care and outcomes. The medical records are reviewed
by centrally trained data abstractors using a standardized and vali-
dated instrument. Each hospital within the MBSC is audited annually

Annals of Surgery r Volume 254, Number 4, October 2011 www.annalsofsurgery.com | 633

Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Finks et al
by nurses from the coordinating center to verify the accuracy and
completeness of its clinical registry data.
For this study, we identified all patients undergoing a primary
(nonrevisional) bariatric surgical procedure between June 2006 and
December 2010. We excluded patients undergoing a revisional pro-
cedure due to the heterogeneity of this patient population and the
increased risk associated with these operations.
Outcomes
Abstracted data include information on 13 different types of
perioperative complications. Complications were determined by doc-
umentation of the specific complication in the chart, including con-
firmatory radiographic imaging reports when available, as well as the
treatment for the complication. Complications were categorized ac-
cording to severity as nonlife-threatening (grade 1), potentially life-
threatening (grade 2), or life-threatening complications associated
with permanent residual disability or death (grade 3). Grade 1 com-
plications included wound infections (requiring antibiotics or opening
of the wound), anastomotic strictures (requiring endoscopic dilation),
bleeding (requiring blood transfusion of 4 units), pneumonia and
hospital acquired infections (urinary tract infection, Clostridium dif-
ficile colitis). Grade 2 complications included abdominal abscess
(requiring percutaneous drainage or reoperation), bowel obstruc-
tion (requiring reoperation), leak (requiring percutaneous drainage
or reoperation), bleeding (requiring blood transfusion > 4 units, en-
doscopy, reoperation or splenectomy), wound infection or dehiscence
(requiring reoperation), respiratory failure (requiring 27 days me-
chanical ventilation), renal failure (requiring in-hospital dialysis),
venous thromboembolism (deep venous thrombosis or pulmonary
embolism) and band-related problems requiring reoperation (port site
infection, gastric perforation, band slippage and outlet obstruction).
Grade 3 complications included myocardial infarction or cardiac ar-
rest, renal failure requiring long-term dialysis, respiratory failure
(requiring > 7 days mechanical ventilation or tracheostomy) and
death. The severity of events which did not clearly meet above criteria
(n = 1855) was determined by the Endpoints Committee made up of
MBSC participating surgeons.
Independent Variables
Data on patient characteristics included demographics (age,
gender, and payer type), height, weight, mobility limitations (requir-
ing ambulation aids, nonambulatory, or bed-bound), smoking status
and comorbid conditions. The height and weight were used to cal-
culate body mass index (BMI), a ratio of the weight in kilograms
to the height in meters squared. The definitions for most comorbidi-
ties included documentation of the condition and its treatment in
the medical record. Comorbid conditions included pulmonary dis-
ease (asthma, obstructive/ restrictive disorders, home oxygen use,
Pickwickian syndrome), cardiovascular disease (coronary artery dis-
ease, dysrhythmia, peripheral vascular disease, stroke, hypertension,
hyperlipidemia), sleep apnea, psychological disorders, prior venous
thromboembolism (VTE), diabetes, chronic renal failure (requiring
dialysis or transplant), liver disease (nonalcoholic fatty liver, cir-
rhosis, liver transplantation), urinary incontinence, gastroesophageal
reflux disease, peptic ulcer disease, cholelithiasis, previous ventral
hernia repair, and musculoskeletal disorders.
Statistical Analysis
Pearson 2 test for categorical variables and one-way analysis
of variance F-test for continuous variables were used for compar-
ing patient characteristics, comorbidities, and complications among
procedure types. The primary outcome variable for this study was
serious complications (grade 2 or 3). Patient characteristics showing
a significant association with serious complications (P < 0.1) were
634 | www.annalsofsurgery.com
Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Annals of Surgery r Volume 254, Number 4, October 2011
entered into a multivariable logistic regression model. Significant risk
factors (P < 0.05) were kept in the model. Probabilities of serious
complications were predicted on the basis of the multivariable model.
There were no missing data for either the primary outcome or the
major covariates, with the exception of pulmonary disease, for which
less than 0.3% of the cases had missing data. Two-way interactions
between procedure types and the main covariates were assessed. The
only significant interaction was between laparoscopic gastric bypass
and gender (P = 0.01). However, the overall fit of the model was bet-
ter without the interaction term and the c-statistic was nearly identical
in models with and without the interaction term. For that reason, we
elected to use a regression model containing only the main effects.
Discrimination of the model was determined using the c-
statistic, or area under the receiver operating characteristic (ROC)
curve.14 This is an estimate of the ability of the model to distinguish
events (serious complications) from nonevents. The c-statistic ranges
from 0.5 to 1.0, where a value of 1.0 represents perfect discrimina-
tion and a value of 0.5 suggests discrimination that is no better than
chance. The calibration of the model was assessed using the Hosmer-
Lemeshow goodness-of-fit test.15 This tests the correlation between
observed and predicted outcomes across the entire range of predicted
outcomes. A low 2 value and high nonsignificant P value for the
Hosmer-Lemeshow test indicates acceptable calibration of the model.
Because predictive models perform better in the data from
which they were derived than on new data,16,17 we performed boot-
strap resampling to obtain bias-corrected confidence intervals and
c-statistic, thereby providing a more honest estimate of model perfor-
mance. We chose this method of internal validation over others, such
as split-sample modeling, because bootstrap resampling techniques
have been shown to produce stable and nearly unbiased estimates
of predictive accuracy, with better efficiency than other methods.17
One hundred random bootstrap samples with replacement, and of
the same size as the original sample were drawn from the original
dataset consisting of all patients. Separate logistic regression models
were developed based on each bootstrap run. The model as estimated
from each bootstrap sample was evaluated in the bootstrap sample
and the original sample using the c-statistic. The difference between
the 2 c-statistics gave an estimate of the optimism in the apparent
performance of the model for that bootstrap run. We then averaged
the optimism across all bootstrap runs to get a stable estimate of the
overall optimism of the model. To obtain the bias-corrected c-statistic,
we subtracted the average estimate of optimism from the c-statistic
obtained using the original model based on the original sample. The
standard error estimates were averaged across the 100 bootstrap sam-
ples to obtain bias-corrected 95% confidence intervals (CIs) for each
regression coefficient.
Using the coefficients obtained from the model, we developed
a risk calculator to estimate an individual patients probability of
developing a serious complication after bariatric surgery. Each risk
factor was used as a dichotomous variable. The risk calculator works
by combining an individual patients values for each risk factor with
the coefficients from the model into a regression equation that yields
the odds of developing a serious complication. From the odds, the
probability of developing a complication is obtained by the following
transformation: probability (%) = 100 odds/(1 + odds). The
regression equation takes the form:
Odds (serious complication) = EXP ( 0 + 1 x1 +
2 x2 + 3 x3 ... + k Xk )
In this equation, EXP is equivalent to ex , where e is the
base of the natural logarithm (2.718); 0 is the model intercept; 1
through k represent the coefficients from the model corresponding
to individual predictors; X1 through Xk represent the patients data for
each risk factor. For items with a yes/no response, yes = 1 and no
= 0. For procedure type, the reference group was adjustable gastric

C 2011 Lippincott Williams & Wilkins
Annals of Surgery r Volume 254, Number 4, October 2011 Predicting Risk for Serious Complications in Bariatric Surgery

band (x = 0). For the remaining procedures, x = 1 for the patients
specific procedure, whereas x is set to 0 for all other procedures.
All analyses were performed using statistical software package
version 9.2 (Cary, NC). The study protocol was approved by the
Institutional Review Boards of the University of Michigan and all
participating hospitals. The requirement for informed consent was
waived.
RESULTS
During the study period, 25,469 patients underwent bariatric
surgery at 29 centers throughout the state of Michigan. Of these,
14,850 (58%) underwent gastric bypass (1092 via an open approach),
8015 (31%) laparoscopic adjustable gastric band, 2279 (9%) sleeve
gastrectomy, and 325 (1%) biliopancreatic diversion with duodenal
switch. The majority of operations (95%) were performed laparo-
scopically and conversion rates were low (1.3%), although 70% of
the duodenal switch procedures were done using an open technique.
Patient characteristics by procedure type are listed in Table 1. Among
patients undergoing bariatric surgery, 78% were women, with a mean
age of 46 years and mean BMI of 48 kg/m2 . There were few substan-
tial differences in patient characteristics among procedures, although
patients undergoing open gastric bypass had the highest mean BMI,
mean number of comorbidities, and rate of mobility limitations.
Rates of individual complications by procedure are listed in
Table 2. Overall, 1245 patients (4.9%) experienced at least 1 minor
complication, 644 patients (2.5%) experienced at least 1 serious com-
plication and 23 patients (0.1%) died after bariatric surgery during
the study period. Rates of serious complications varied substantially
across procedures, ranging from 0.9% with adjustable gastric banding
to 8% with the duodenal switch procedure, with mortality rates be-
tween 0.04% for adjustable gastric banding and 0.5% for open gastric
bypass.
Significant risk factors for serious complications on multivari-
able analysis are detailed in Table 3. The most significant risk factor
was procedure type, with duodenal switch carrying the highest risk
for serious complications (odds ratio [OR] 9.68; CI 6.0515.49), rela-
tive to adjustable gastric band (reference group). The risk was similar
for laparoscopic (OR 3.58; CI 2.794.64) and open (OR 3.51; CI
2.385.22) gastric bypass and somewhat less for sleeve gastrectomy
(OR 2.46; CI 1.733.50). Significant patient risk factors included pre-
vious history of VTE (OR 1.90; CI 1.412.54), mobility limitations
(OR 1.61; CI 1.232.13), coronary artery disease (OR 1.53; CI 1.17
2.02), age over 50 years (OR 1.38; CI 1.181.61), pulmonary disease
(OR 1.37; CI 1.151.64), male gender (OR 1.26; CI 1.061.50), and
smoking history (OR 1.20; CI 1.021.40). The model demonstrated
moderate discrimination (c-statistic 0.68; bias-corrected c-statistic
0.66) and good calibration (Hosmer-Lemeshow test 2 = 10.82, P =
0.2119). Figure 1 shows the calibration curve for the risk prediction
model across deciles of predicted risk.
Table 3 details the risk prediction equation based on the co-
efficients from the final regression model. On the basis of this risk
prediction model, 92% of patients had a predicted risk of serious com-
plications less than 5%. The predicted risk for serious complications
was between 5% and 10% for 7% of patients and greater than 10% in
1% of patients. A user-friendly version of this risk calculator will be
available on the MBSC website at https://www.michiganbsc.org. As
an example of how to use the risk calculator: a 60-year-old woman
with a history of smoking and chronic lung disease, but no additional
comorbidities, is scheduled to undergo laparoscopic gastric bypass.
Her estimated risk of developing a serious complication is 4.6%.

TABLE 1. Characteristics of Patients Undergoing Bariatric Surgery in Michigan Between June 2006 and December 2010
Lap Gastric Open Gastric Adjustable Sleeve Duodenal
Patient Overall Bypass Bypass Gastric Band Gastrectomy Switch
Characteristics N = 25,469 N = 13,758 N = 1092 N = 8015 N = 2279 N = 325 P
Demographics/Other Characteristics
Mean age ( SD) 45.7 (11.4) 45.3 (11.1) 46.5 (11.1) 46.6 (12.0) 45.4 (11.5) 43.1 (10.2) <0.0001
Male gender, no. (%) 5525 (21.7) 2810 (20.4) 283 (25.9) 1706 (21.3) 671 (29.4) 55 (16.9) <0.0001
Private insurance no. (%) 18,333 (72.0) 9383 (68.2) 680 (62.3) 6169 (77.0) 1931 (84.7) 170 (52.3) <0.0001
Mean BMI* ( SD) 48.0 (8.5) 48.6 (8.2) 53.5 (11.3) 44.9 (6.9) 52.0 (9.5) 51.9 (8.5) <0.0001
Any smoking history no. (%) 9919 (39.0) 5623 (40.9) 431 (39.5) 2881 (36.0) 843 (37.0) 141 (43.4) <0.0001
Mobility limitations no. (%) 1360 (5.3) 717 (5.2) 100 (9.2) 366 (4.6) 158 (6.9) 19 (5.9) <0.0001
Comorbidities, no. (%)
History of previous VTE 979 (3.8) 540 (3.9) 62 (5.7) 270 (3.4) 103 (4.5) 4 (1.2) <0.0001
Pulmonary disease 6588 (25.9) 3607 (26.3) 461 (42.3) 1901 (23.8) 508 (22.4) 111 (34.4) <0.0001
Sleep apnea 11,374 (44.8) 6356 (46.3) 614 (56.3) 3081 (38.5) 1159 (51.0) 164 (50.5) <0.0001
Hypertension 13,544 (54.9) 7290 (54.8) 692 (64.6) 4139 (53.3) 1268 (56.9) 155 (49.1) <0.0001
Coronary artery disease 1369 (5.4) 727 (5.3) 73 (6.7) 445 (5.6) 113 (5.0) 11 (3.4) 0.103
Peripheral vascular disease 367 (1.4) 203 (1.5) 25 (2.3) 103 (1.3) 34 (1.5) 2 (0.6) 0.071
Diabetes 8540 (33.7) 4856 (35.5) 459 (42.2) 2367 (29.7) 752 (33.2) 106 (32.7) <0.0001
Hyperlipidemia 12,774 (50.2) 7127 (51.8) 642 (58.8) 3835 (47.9) 1070 (47.0) 100 (30.8) <0.0001
Chronic renal failure 57 (0.2) 14 (0.1) 2 (0.2) 35 (0.4) 6 (0.3) 0 <0.0001
Liver disease 1091 (4.3) 594 (4.3) 144 (13.2) 271 (3.4) 74 (3.3) 8 (2.5) <0.0001
Urinary incontinence 5619 (22.1) 3126 (22.7) 251 (23.0) 1565 (19.5) 526 (23.1) 151 (46.5) <0.0001
Gastroesophageal reflux disease 12,315 (48.4) 6,691 (48.6) 683 (62.6) 3727 (46.5) 995 (43.7) 219 (67.4) <0.0001
Peptic ulcer disease 716 (2.8) 446 (3.2) 43 (3.9) 147 (1.8) 71 (3.1) 9 (2.8) <0.0001
Musculoskeletal disorder 19,714 (77.4) 11,034 (80.2) 975 (89.3) 5826 (72.7) 1601 (70.3) 278 (85.5) <0.0001
Cholelithiasis 734 (2.9) 360 (2.6) 139 (12.7) 124 (1.6) 73 (3.2) 38 (11.7) <0.0001
Previous ventral hernia repair 714 (2.8) 350 (2.5) 74 (6.8) 178 (2.2) 102 (4.5) 10 (3.1) <0.0001
Mean number of comorbidities ( SD) 4.4 (2.1) 4.6 (2.1) 5.5 (2.0) 4.1 (2.1) 4.3 (2.2) 5.0 (2.0) 0.001
*Body mass index (kg/m2 ).
Venous thromboembolism (deep venous thrombosis or pulmonary embolism).
Includes stroke, transient ischemic attack, abdominal aortic aneurysm, and lower extremity arterial occlusive disease.


C 2011 Lippincott Williams & Wilkins www.annalsofsurgery.com | 635

Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Finks et al Annals of Surgery r Volume 254, Number 4, October 2011

TABLE 2. Unadjusted Rates of Complications Among Patients Undergoing Bariatric Surgery in Michigan Between June 2006
and December 2010
Lap Gastric Open Gastric Adjustable Sleeve Duodenal
Overall Bypass Bypass Gastric Band Gastrectomy Switch
N = 25,469 N = 13,758 N = 1092 N = 8015 N = 2279 N = 325 P
Minor complications, no. (%)
Wound problem* 563 (2.2) 370 (2.7) 65 (5.95) 82 (1.0) 30 (1.3) 16 (4.9) <0.0001
Anastomotic stricture 281 (1.1) 249 (1.8) 13 (1.2) 0 15 (0.7) 4 (1.2) <0.0001
Minor bleeding 335 (1.3) 281 (2.0) 24 (2.2) 11 (0.1) 16 (0.7) 3 (0.9) <0.0001
Pneumonia 196 (0.8) 133 (1.0) 15 (1.4) 23 (0.3) 13 (0.6) 12 (3.7) <0.0001
Urinary tract infection 72 (0.3) 50 (0.4) 7 (0.6) 10 (0.1) 4 (0.2) 1 (0.3) <0.0001
C. difficile 19 (0.1) 12 (0.1) 0 1 6 (0.3) 0 0.0007
Number of patients with 1 minor 1245 (4.9) 932 (6.8) 102 (9.3) 109 (1.4) 75 (3.3) 27 (8.3) <0.0001
complications
Serious complications, no. (%)
Abdominal abscess 116 (0.5) 69 (0.5) 5 (0.5) 7 (0.1) 21 (0.9) 14 (4.3) <0.0001
Leak 151 (0.6) 109 (0.8) 7 (0.6) 0 22 (1.0) 13 (4.0) <0.0001
Bowel obstruction 111 (0.4) 105 (0.8) 1 (0.1) 1 (0.01) 3 (0.1) 1 (0.3) <0.0001
Major bleeding 118 (0.5) 94 (0.7) 13 (1.2) 3 (0.04) 7 (0.3) 1 (0.3) <0.0001
Respiratory Failure 85 (0.3) 63 (0.5) 5 (0.5) 3 (0.04) 6 (0.3) 8 (2.5) <0.0001
Renal failure 56 (0.2) 44 (0.3) 4 (0.4) 4 (0.05) 2 (0.1) 1 (0.3) 0.0005
Wound problem* 77 (0.3) 37 (0.3) 15 (1.4) 2 (0.02) 3 (0.1) 15 (4.6) <0.0001
Venous thromboembolism 87 (0.3) 55 (0.4) 8 (0.7) 8 (0.1) 10 (0.4) 5 (1.5) <0.0001
Myocardial infarction/cardiac arrest 22 (0.1) 15 (0.1) 4 (0.4) 1 (0.01) 1 (0.04) 1 (0.3) 0.0012
Band-related complications 47 (0.4)
Death 23 (0.1) 12 (0.1) 5 (0.5) 3 (0.04) 2 (0.1) 1 (0.3) 0.0004
Number of patients with 1 serious 644 (2.5) 450 (3.3) 40 (3.7) 75 (0.9) 53 (2.3) 26 (8.0) <0.0001
complications
*Includes infection, hematoma and dehiscence. Minor wound problems: requiring antibiotics and/or opening of the wound; Serious wound problems: requiring reoperation.
Minor bleeding: requiring transfusion of 4 units PRBCs; Serious bleeding: requiring transfusion of >4 units PRBCs, endoscopy, reoperation or splenectomy).
Requiring reoperation. Includes reoperation for band slippage, outlet obstruction, gastric/esophageal perforation, and port site infection.

TABLE 3. Significant Risk Factors for Serious Complications on Multivariable Analysis Among Patients Undergoing
Bariatric Surgery in Michigan Between June 2006 and December 2010
Risk Factor Regression Coefficient Odds Ratio 95% CI Bias-Corrected 95% CI* P*
Procedure type (Ref. Adjustable gastric band)
Duodenal switch 2.2702 9.68 6.0815.41 6.0515.49 0.0001
Laparoscopic gastric bypass 1.2759 3.58 2.804.58 2.794.64 0.0001
Open gastric bypass 1.2556 3.51 2.375.20 2.385.22 0.0001
Sleeve gastrectomy 0.8988 2.46 1.723.51 1.733.50 0.0001
Patient factors
Previous venous thromboembolism 0.6410 1.90 1.422.54 1.412.54 0.0001
Mobility limitations 0.4784 1.61 1.242.10 1.232.13 0.0004
Coronary artery disease 0.4260 1.53 1.162.02 1.172.02 0.002
Age over 50 years 0.3225 1.38 1.171.63 1.181.61 0.0001
Pulmonary disease 0.3150 1.37 1.161.62 1.151.64 0.0003
Male gender 0.2321 1.26 1.051.51 1.061.50 0.01
Smoking history 0.1797 1.20 1.021.40 1.021.40 0.03
*Derived from bias-corrected estimates of standard error using bootstrap resampling.
Use patient values and coefficients from the model to calculate odds of developing a serious complication using the equation below. The probability (%) =
odds/(1 + odds). The notation EXP is equivalent to ex , where e is the base of the natural logarithm (2.718). For items with a yes/no response, yes = 1 and no = 0.
For procedure type, the reference group is adjustable gastric band.
Odds = EXP ([5.12] + 2.2702 [duodenal switch] + 1.2759 [laparoscopic gastric bypass] + 1.2556 [open gastric bypass] + 0.8988 [sleeve
gastrectomy] + 0 [adjustable gastric band] + 0.6410 [previous VTE] + 0.4784 [mobility] + 0.4260 [coronary artery disease] + 0.3225 [age over 50] +
0.3150 [pulmonary disease] + 0.2321 [male gender] + 0.1797 [any smoking history]).

DISCUSSION trectomy and adjustable gastric band, in descending order. We iden-

In this analysis of patients undergoing bariatric surgery in
Michigan, the overall rate of mortality (0.1%) and serious compli-
cations (2.5%) were low but varied considerably by procedure type.
The probability of serious complications was highest for patients
undergoing duodenal switch, followed by gastric bypass, sleeve gas-
tified several patient factors which contributed to the risk for serious
complications, the most significant being a prior history of VTE. Ad-
ditional risk factors which were significant on multivariate analysis
included mobility limitations, coronary artery disease, age over 50
years, pulmonary disease, male gender, and any smoking history.

636 | www.annalsofsurgery.com 
C 2011 Lippincott Williams & Wilkins

Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Annals of Surgery r Volume 254, Number 4, October 2011
FIGURE 1. Calibration plot for the prediction model (n =
25,469). Observed versus predicted probability of 30-day seri-
ous complications across deciles of predicted risk. Dashed line
indicates perfect calibration.
The rate of serious complications in our study is consistent with
that of other large contemporaneous series (3.7%4.3%).12,18 Further-
more, several of the risk factors for serious complications identified
in our study have been reported previously. Age, male gender, prior
VTE, pulmonary disease, cardiac disease and smoking have all been
associated with risk of morbidity after bariatric surgery.12,1824 Al-
though mobility limitations per se has not been previously associated
with morbidity from bariatric surgery, 2 recent studies identified de-
pendent functional status as a risk factor for serious complications and
it is likely that these 2 risk factors are related.12,18 Notably, body mass
index, a commonly reported risk factor in other studies of bariatric
surgery,12,18,19,21 was not associated with risk for serious complica-
tions in our study.
As with our study, previous reports have also documented dif-
ferential rates of complications among bariatric procedures. Restric-
tive procedures, such as the laparoscopic adjustable gastric band, have
generally proven to be the safest of the bariatric procedures.1,12,2527
There is a relative paucity of data on sleeve gastrectomy, but most
studies support our finding that sleeve gastrectomy is a safer alter-
native to gastric bypass or duodenal switch.2830 In our study the
duodenal switch procedure was strongly associated with risk for seri-
ous complications and other studies have demonstrated higher com-
plication rates with this operation when compared to other bariatric
procedures.1,12,31 However, the overall number of these cases was low
in our series and nearly 3/4 of these were performed using an open
approach. Our model with regard to duodenal switch may not be
applicable in regions where these procedures are done with greater
frequency and using a laparoscopic approach.32
The MBSC maintains one of the largest prospective and exter-
nally audited population-based registries of bariatric surgery patients.
As such there is a clinically rich dataset well-suited for developing pre-
diction models not only for complications but also for other outcomes
such as weight loss and comorbidity resolution. Two recent studies
analyzed data from the National Surgical Quality Improvement Pro-
gram (NSQIP) to develop predictive models for major morbidity
in bariatric surgery.12,18 Turner and colleagues examined data from
over 30,000 bariatric surgery patients and identified age, BMI, serum
albumin and functional status as significant predictors.18 Gupta,
et al, reviewed data from over 11,000 patients and found that recent
MI/angina, dependent functional status, bleeding disorder, hyperten-
sion, BMI and procedure type were most associated with serious
complications.12 NSQIP has the advantage of access to a large and
nationally-representative data source. However, NSQIP was designed
to capture data on a wide spectrum of procedures and lacks specificity
with regard to bariatric-specific comorbidities (eg, previous VTE) and

C 2011 Lippincott Williams & Wilkins
Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Predicting Risk for Serious Complications in Bariatric Surgery
outcomes (eg, anastomotic leak). Furthermore, patients undergoing
sleeve gastrectomy were not included in the NSQIP database.
It should be noted that the discrimination of our prediction
model (ie, the ability to separate cases from noncases) was only mod-
erate, with a c-statistic of 0.68 (bias-corrected to 0.66), which was
similar to that with the studies using NSQIP data.12,18 This may be due,
to some extent, to the relative homogeneity of the patient population.
Given the elective nature of bariatric surgery, the highest risk patients
may be selected out. This would be in contrast to cardiac surgery,
where predictive models from the large Society of Thoracic Surgery
National Adult Cardiac Surgery Database demonstrate discrimination
indices over 0.70.33 Another possible reason for the somewhat limited
discrimination of the model is that factors other than patient charac-
teristics, including surgeon volume10,3436 and technical variation,37
may also influence surgical outcomes. Finally, serious complications
encompass a somewhat heterogeneous group of outcomes and one
would expect that risk factors significant for some complications may
not predict others.
Given the size and clinical richness of our dataset, however, this
paper likely represents the best available evidence on patient-level risk
factors for serious complications after bariatric surgery and can still
be of benefit to patients and providers. First, some of these risk fac-
tors can be addressed directly to minimize their impact. For example,
patients with prior history of VTE may benefit from more aggres-
sive VTE prophylaxis, such as postdischarge chemoprophylaxis.38
Patients with underlying cardiopulmonary disease should have these
conditions optimized preoperatively, whereas those with mobility lim-
itations may benefit from prehabilitation,39 as a means of enhancing
their functional capacity before surgery.
Our risk prediction model offers additional benefits as well.
First, it will serve as a useful adjunct in the informed consent process
by allowing providers to give patients a more accurate estimate of
their individual risk with bariatric surgery. This tool will also help
identify patients at particularly high risk for complications, for whom
lower-risk procedures, such as sleeve gastrectomy or adjustable gas-
tric band, may be preferable. This prediction model will also improve
the process of risk-adjustment for use in comparing different proce-
dures or assessing interventions to reduce complications. Ultimately,
more accurate accounting of patient risk with bariatric surgery will
better inform coverage and approval decisions by Medicare and other
health insurers.
Our study does have several limitations. First, the registry does
not capture events occurring beyond 30 days after the bariatric proce-
dure, unless the patient remains an inpatient or is readmitted before
the 30 day window. This could lead to under-reporting of the rate
of serious complications. However, most serious complications are
likely to present within the first 30 days of the procedure and it is un-
likely that these uncaptured events would have a significant impact on
the association with risk factors identified in the study. An additional
potential concern is that our analysis does not account for surgeon
volume, a well-recognized predictor of outcome.3436 However, we
felt that this was appropriate, as our goal was to develop a model of
risk prediction based on patient-level variables, which could be used
across all providers. Adding provider volume to the model would
make it impractical for use by surgeons. Finally, even with the use
of bootstrapping to obtain bias-adjusted odds ratios and confidence
intervals, our risk prediction model may not perform as well in pa-
tients undergoing bariatric surgery outside of Michigan. The MBSC
is a quality improvement initiative unique to Michigan, whose goal
is to improve outcomes with bariatric surgery across all programs.
It is certainly possible that participation in the MBSC could alter
the relationship between patient risk factors and outcomes. Further
studies are needed to validate this prediction model in an independent
population of bariatric surgery patients.
www.annalsofsurgery.com | 637
Finks et al
In conclusion, we have developed a predictive model for se-
rious complications after bariatric surgery. It was developed from
patient data across a diverse group of hospitals in Michigan, which
include academic and nonacademic centers, large and small, urban,
suburban and rural. The model, therefore, has wide applicability. Cer-
tainly, the predictive ability of the model is not ideal at the patient
level and is not meant for high-stakes provider profiling (as in Pay
for Performance or other quality-based reimbursement programs).
However, we believe that it provides a best estimate of patient risk
and we expect that it will prove valuable to patients and providers.
Furthermore, it will improve risk stratification for use in assessment
of new procedures and techniques to improve outcomes for bariatric
surgery patients.
REFERENCES
1. Buchwald H, Avidor Y, Braunwald E, et al. Bariatric surgery: a systematic
review and meta-analysis. JAMA. 2004;292:17241737.
2. Sjostrom L, Lindroos AK, Peltonen M, et al. Lifestyle, diabetes, and car-
diovascular risk factors 10 years after bariatric surgery. N Engl J Med.
2004;351:26832693.
3. Dixon JB, OBrien PE, Playfair J, et al. Adjustable gastric banding and con-
ventional therapy for type 2 diabetes: a randomized controlled trial. JAMA.
2008;299:316323.
4. Adams TD, Gress RE, Smith SC, et al. Long-term mortality after gastric bypass
surgery. N Engl J Med. 2007;357:753761.
5. Christou NV, Sampalis JS, Liberman M, et al. Surgery decreases long-term
mortality, morbidity, and health care use in morbidly obese patients. Ann Surg.
2004;240:416423; discussion 2324.
6. Sjostrom L, Narbro K, Sjostrom CD, et al. Effects of bariatric surgery on
mortality in Swedish obese subjects. N Engl J Med. 2007;357:741752.
7. Buchwald H, Oien DM. Metabolic/bariatric surgery Worldwide 2008. Obes
Surg. 2009;19:16051611.
8. Santry HP, Gillen DL, Lauderdale DS. Trends in bariatric surgical procedures.
JAMA. 2005;294:19091917.
9. Encinosa WE, Bernard DM, Du D, et al. Recent improvements in bariatric
surgery outcomes. Med Care. 2009;47:531535.
10. Birkmeyer NJ, Dimick JB, Share D, et al. Hospital complication rates with
bariatric surgery in Michigan. JAMA. 304:435442.
11. Flum DR, Belle SH, King WC, et al. Perioperative safety in the longitudinal
assessment of bariatric surgery. N Engl J Med. 2009;361:445454.
12. Gupta PK, Franck C, Miller WJ, et al. Development and validation of a bariatric
surgery morbidity risk calculator using the prospective, multicenter NSQIP
dataset. J Am Coll Surg. 212:301309.
13. Birkmeyer NJ, Birkmeyer JD. Strategies for improving surgical qualityshould
payers reward excellence or effort? N Engl J Med. 2006;354:8648670.
14. Swets JA. Measuring the accuracy of diagnostic systems. Science.
1988;240:12851293.
15. Hosmer DW, Lemeshow S. Applied Logistic Regression. New York: John Wiley
& Sons, Inc., 1989.
16. Bleeker SE, Moll HA, Steyerberg EW, et al. External validation is necessary in
prediction research: a clinical example. J Clin Epidemiol. 2003;56:826832.
17. Steyerberg EW, Harrell FE Jr., Borsboom GJ, et al. Internal validation of pre-
dictive models: efficiency of some procedures for logistic regression analysis.
J Clin Epidemiol. 2001;54:774781.
18. Turner PL, Saager L, Dalton J, et al. A Nomogram for predicting surgical
complications in bariatric surgery patients. Obes Surg. 2011;21:655662.
19. DeMaria EJ, Portenier D, Wolfe L. Obesity surgery mortality risk score: pro-
posal for a clinically useful score to predict mortality risk in patients undergoing
gastric bypass. Surg Obes Relat Dis. 2007;3:134140.
20. Fernandez AZ Jr., DeMaria EJ, Tichansky DS, et al. Experience with over
3,000 open and laparoscopic bariatric procedures: multivariate analysis of
factors related to leak and resultant mortality. Surg Endosc. 2004;18:193197.
21. Livingston EH, Arterburn D, Schifftner TL, et al. National surgical qual-
ity improvement program analysis of bariatric operations: modifiable risk
factors contribute to bariatric surgical adverse outcomes. J Am Coll Surg.
2006;203:625633.
22. Poulose BK, Griffin MR, Zhu Y, et al. National analysis of adverse patient
safety for events in bariatric surgery. Am Surg. 2005;71:406413.
638 | www.annalsofsurgery.com
Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Annals of Surgery r Volume 254, Number 4, October 2011
23. Weller WE, Rosati C, Hannan EL. Predictors of in-hospital postoperative com-
plications among adults undergoing bariatric procedures in New York state,
2003. Obes Surg. 2006;16:702708.
24. Livingston EH, Huerta S, Arthur D, et al. Male gender is a predictor of mor-
bidity and age a predictor of mortality for patients undergoing gastric bypass
surgery. Ann Surg. 2002;236:576582.
25. Lindsey ML, Patterson WL, Gesten FC, et al. Bariatric surgery for obesity:
surgical approach and variation in in-hospital complications in New York State.
Obes Surg. 2009;19:688700.
26. Nguyen NT, Slone JA, Nguyen XM, et al. A prospective randomized trial of
laparoscopic gastric bypass versus laparoscopic adjustable gastric banding for
the treatment of morbid obesity: outcomes, quality of life, and costs. Ann Surg.
2009;250:631641.
27. Parikh MS, Laker S, Weiner M, et al. Objective comparison of complications re-
sulting from laparoscopic bariatric procedures. J Am Coll Surg. 2006;202:252
261.
28. Gumbs AA, Gagner M, Dakin G, Pomp A. Sleeve gastrectomy for morbid
obesity. Obes Surg. 2007;17:962969.
29. Rosen DJ, Dakin GF, Pomp A. Sleeve gastrectomy. Minerva Chir.
2009;64:285295.
30. Tucker ON, Szomstein S, Rosenthal RJ. Indications for sleeve gastrectomy as
a primary procedure for weight loss in the morbidly obese. J Gastrointest Surg.
2008;12:662667.
31. Buchwald H, Estok R, Fahrbach K, et al. Trends in mortality in bariatric
surgery: a systematic review and meta-analysis. Surgery. 2007;142:621632;
discussion 3235.
32. Sovik TT, Taha O, Aasheim ET, et al. Randomized clinical trial of laparoscopic
gastric bypass versus laparoscopic duodenal switch for superobesity. Br J Surg.
97:160166.
33. Shahian DM, OBrien SM, Filardo G, et al. The Society of Thoracic Surgeons
2008 cardiac surgery risk models: part 1coronary artery bypass grafting
surgery. Ann Thorac Surg. 2009;88:S222.
34. Campos GM, Ciovica R, Rogers SJ, et al. Spectrum and risk factors of com-
plications after gastric bypass. Arch Surg. 2007;142:969975; discussion 76.
35. Courcoulas A, Schuchert M, Gatti G, et al. The relationship of surgeon and
hospital volume to outcome after gastric bypass surgery in Pennsylvania: a
3-year summary. Surgery. 2003;134:613621; discussion 2123.
36. Flum DR, Salem L, Elrod JA, et al. Early mortality among Medicare benefi-
ciaries undergoing bariatric surgical procedures. JAMA. 2005;294:19031908.
37. Finks JF, Carlin A, Share D, et al. Effect of surgical techniques on clinical
outcomes after laparoscopic gastric bypass-results from the Michigan Bariatric
Surgery Collaborative. Surg Obes Relat Dis. 2011;7:284289.
38. Raftopoulos I, Martindale C, Cronin A, et al. The effect of extended
post-discharge chemical thromboprophylaxis on venous thromboembolism
rates after bariatric surgery: a prospective comparison trial. Surg Endosc.
2008;22:23842391.
39. Carli F, Charlebois P, Stein B, et al. Randomized clinical trial of prehabilitation
in colorectal surgery. Br J Surg. 97:11871197.
DISCUSSANTS
A. Forse (Omaha, NE):
Dr. Finkss analysis uses a rich and well-defined database from
the Michigan Bariatric Collaborative, and through appropriate statis-
tics, is able to identify risk factors and to generate a risk calculator
for patient and physician use. These data are consistent with a study
our group recently reported using the American College of Surgeons
NSQIP database.
I have several questions. First is that a database is only as good
as the accuracy of the data entered. How did the Michigan Bariatric
Collaborative control the data input? What definitions were used for
the specific complications and how did you assure the accuracy of
the data input by the nurse reviewers? With the NSQIP database,
the nurse reviewers are audited on a regular basis and their data
must be below a critical threshold to be used. What did you do
when there were data inaccuracies? Although clinically useful, do
you think that we should be using BMI to measure obesity when
performing these statistical exercises? Do you think that your database
is a representative sampling in that you used patients operated on in
Michigan and by Michigan surgeons resulting perhaps in a sequenced

C 2011 Lippincott Williams & Wilkins
Annals of Surgery r Volume 254, Number 4, October 2011
sampling? Does this limit its applicability in the rest of the United
States or elsewhere?
Second, how did you control for the interactions between the
types of surgery and the other variables studied, as such an important
interaction can affect outcomes? You make mention of this problem
in the paper, but it is not clear on how you addressed it. For exam-
ple, in our analysis of the NSQIP database, we placed all second-
order interaction terms involving procedures into the list of candidate
variables.
The surgical procedure problem is also an issue with the BPD
operations, which were predominantly open. This factor would have
increased the complication risk of this operation unfairly, as it is
known that open bariatric surgery is more complicated than laparo-
scopic. Perhaps equally so is the case with gastric banding, whose
laparoscopic complications are less than the open banding proce-
dures. Should this be an independent variable?
I agree with your statistical methods; however, I am not sure,
with such a robust and representative database being collected, why
you chose to use the bootstrapping technique and limit the sampling to
only 100. The bootstrapping techniques, using the database in which
the modeling was done to test the model, still has an inherent bias,
does not provide general finite sample guarantees, and has a tendency
to be overly optimistic. Why not use the modeling prospectively on
new patients collected to test the validity? In the time since you wrote
the abstract, you should have had enough patients to test the model.
Is it holding up?
What is very interesting is that both database studies, yours
and ours, generated virtually identical c-statistics of 0.66. Why do
you think this is the case and why were these predictive models using
such large databases not able to achieve better predictive power?
Finally, do you propose to use the calculator to monitor individ-
ual surgeons performance? Should you use the observed-to-expected
complication ratio for credentialing or can you use it to learn from
surgeons who seem to have outstanding outcomes, which could then
be shared and tried out within your collaborative?
Response from J. F. Finks
As far as definitions for the complications, we had very specific
definitions that included documentation in the medical record as well
as imaging findings. An important part of the definition was the
treatment that was required for the complication, because that is what
determined the severity of the complication.
Certainly, data input errors are an issue. We did extensive
training of data abstractors initially and have an ongoing relationship
with all of them. We do training at each of our meetings, which occur
3 times per year. In addition, our nurse coordinators are in frequent
communication with the data abstracters to help correct mistakes and
clarify ambiguous issues.
When there are data inaccuracies, we have no trouble going
back to the original site and correcting those inaccuracies. Further-
more, each site in the MBSC is audited annually by nurses from the
data coordinating center to confirm the accuracy of its registry data.
And for events that are difficult to categorize, an endpoints committee
made up of MBSC surgeons helps to define these complications.
Regarding the use of body mass index, it does lack sensitivity
as a measure of adiposity, particularly in men. However, BMI has
been shown to correlate with obesity-related health outcomes, and
in our own work, it has been shown to correlate with venous throm-
boembolism after bariatric surgery. I believe it is a valid measure for
use in risk prediction models.
The question of the representative nature of our collaborative
is a very good one. Certainly, it represents all patients in the state of
Michigan and includes results from a broad range of hospitals, in-

C 2011 Lippincott Williams & Wilkins
Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Predicting Risk for Serious Complications in Bariatric Surgery
cluding urban and rural, large and small, academic and nonacademic.
Its a very large database, and, in that sense, representative.
However, we are part of a very unique collaborative and have
been so for the last several years, with ongoing quality improvement
efforts, which obviously affect complications across all participating
centers. I think it is very important to externally validate our model
with other data sources, such as that through the BOLD database,
the American College of Surgeons or the Longitudinal Assessment
of Bariatric Surgery.
Regarding your question on variable interactions, we did ex-
amine interactions between procedure type and all other significant
risk factors in the regression model. There was a significant inter-
action between gastric bypass and gender. However, including the
interaction term did not improve the model fit and we therefore chose
to use the main effects model only.
Regarding your question with the BPD, certainly, we did assess
the model taking BPD out of the model, and it did almost nothing to
change the coefficients of the model, probably because there are so
few BPD patients. We also collapsed gastric bypass to include both
open and laparoscopic cases and then included operative approach,
open versus laparoscopic, in the model and found that it was not
a significant predictor of serious complications. And certainly, our
model may not be applicable to duodenal switch in settings where it
is done with greater frequency, using a laparoscopic approach.
Regarding your question for gastric banding, we actually had
no open bands placed in our study, and I think the majority of these
are done laparoscopically.
Regarding your question for assessment of internal validity of
the model, bootstrapping resampling techniques have been shown to
provide stable and nearly unbiased estimates of predictive accuracy,
although internal validation is really no substitute for external vali-
dation. Your idea of validating our model with future patients is an
excellent one, and one that we will definitely pursue. At this time,
we do not have clean data on a large enough sample of new patients
to adequately validate the model. Regarding the number of bootstrap
runs, there is no clear evidence that suggests what the optimal num-
ber of bootstrap runs is. I think, given our large sample size, adding
additional bootstrap runs would not improve the internal validation
process.
You bring up an interesting point about why our model and
your model and the other study that used NSQIP data have relatively
moderate discrimination. And I think there are several reasons for
this. First, bariatric surgery patients are, in a sense, a homogeneous
group. Certainly, the highest risk patients do not get operated on
for this elective procedure. In addition, serious complications are a
fairly heterogeneous group of complications, so a risk factor that
is significant for VTE may not be significant for anastomotic leak.
Finally, there are other factors that contribute to outcomes of bariatric
surgery, such as provider volume and techniques used. And I think
that that also contributes to the relatively low discrimination of the
models.
In terms of how we use the model, we certainly do use the risk
prediction model to risk stratify when profiling providers and hospi-
tals for feedback to their own centers. I would not use this for a high
stakes provider or hospital profiling, as with pay-for-performance
reimbursement because the discrimination is really not optimal.
DISCUSSANTS
P. Schauer (Cleveland, OH):
I was surprised to learn from your study that several factors
recognized by other studies did not correlate with operative risk; for
example, body mass index. The Longitudinal Assessment of Bariatric
www.annalsofsurgery.com | 639
Finks et al
Surgery (LABS) study supported by NIH, as well as work by Eric
Demaria and others, have shown a clear relationship between body
mass index and operative mortality and morbidity. Could you explain
why there was not relationship between BMI and complications in
your study? Some insurance companies dont allow an operation to
be performed for patients that have extremely high BMIs because of
the presumed risk. Is it possible that there were not enough patients
with high BMI in your sample?
Other studies have also identified obstructive sleep apnea as
a major predictor, but yours does not. Might that be a matter of
inadequate recognition of sleep apnea among the various centers in
your study?
You did not track the surgeon volume, but what about sur-
geon experience? Most other studies have identified the surgeon life-
time experience before the study as a factor in predicting overall
outcomes.
One final question: Over this 4-year period, did you see an ac-
tual decrease or stabilization in the overall complication rates among
the centers?
Response from J. F. Finks
We looked at body mass index as a categorical variable and
as a continuous variable, but it did not predict serious complications
in our model. However, body mass index was a significant predictor
of certain specific complications, such as venous thromboembolism.
But this is a heterogeneous group of complicationsmany of these
were leaks and abdominal abscessesmany of which do not correlate
well with BMI, which may explain why BMI was not significant in
our model.
We did have a significant number of high BMI patients, as
just under 10% had a BMI over 60. And we did, of course, include
sleep apnea as a risk factor in the model, but it proved to be not
significant. The overall rate of sleep apnea was 45%, which may
represent underreporting because other studies have reported rates up
to 80%.
Regarding your question on surgeon volume, several studies
have shown that surgeon volume is an important predictor of outcomes
with bariatric surgery. And we would not argue that. However, we did
not include provider volume in our model because our objective was to
develop a patient-level risk calculator that could be used by providers
across all hospitals to help them better inform their patients about
surgical risk and identify problems that should be addressed in the
perioperative period.
Interestingly, we have seen a decline in overall complication
rates because the inception of the MBSC. However, there has not
640 | www.annalsofsurgery.com
Copyright 2011 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Annals of Surgery r Volume 254, Number 4, October 2011
been a substantial decline in serious complications during that time.
Currently we are targeting several of the serious complications, such
as venous thromboembolism, in our quality improvement efforts.
DISCUSSANTS
H. Buchwald (Minneapolis, MN):
I have 3 questions. First, do you or your group use your algo-
rithm to judge a patient preoperatively with respect to relative risk?
And if so, do you share this assessment with the patient? Finally, do
you think your analysis is going to increase or minimize the difficul-
ties carriers have placed on bariatric surgery?
Response from J. F. Finks
Currently, we do not use our algorithm for predicting risk for
serious complications. We are in the process of developing this model
for placement on our Web site. We hope that it will prove a useful
tool for providers to better inform their patients about their individual
risk with bariatric surgery.
It is hard to predict how insurance providers will use the infor-
mation. I hope they do not use the risk prediction model to prevent
their large, higher-risk patients from getting bariatric surgery. The
health benefits of bariatric surgery for these patients are substantial,
and it would be wrong to deny them the benefits of surgery simply
because they carry higher risk for complications.
DISCUSSANTS
B. Wolfe (Portland, OR):
As I understand from the paper, your method of capture of com-
plications is the hospital chart. About half of postoperative mortality
in bariatric surgery occurs among patients who have been discharged
from the hospital. And some of these do not return to the index hos-
pital because of the regionalization of bariatric surgery. Some die
without going back to any hospital.
How can we be assured that you have complete capture of these
complications?
Response from J. F. Finks
We probably dont have complete capture of complications.
Certainly patients who develop a complication after discharge and are
treated in a different hospital than the index site may not be captured.
However, we work very closely with the participating hospitals and
surgeons to ensure that such events are reported to the data abstractor
for the index hospital or to our coordinating center when they are
discovered.

C 2011 Lippincott Williams & Wilkins